fix(ci): fix GHA build errors and tests

- Support CRAYON_FORCE_CUDA/ROCM in setup.py for CI wheel building
- Corrected CPU tokenization test to load profile first
- Added python3-dev dependency to CUDA/ROCm build containers

Author: Electroiscoding

.github/workflows/production.yml

@@ -52,6 +52,7 @@ jobs:
           python -c "
           from crayon import CrayonVocab
           v = CrayonVocab(device='cpu')
+          v.load_profile('lite') # LOAD PROFILE FIRST
           result = v.tokenize('Hello Cloud! Testing CRAYON on GitHub Actions.')
           print(f'✅ Tokenized to {len(result)} tokens')
           print(f'   Tokens: {result[:10]}...')
@@ -91,7 +92,7 @@ jobs:
       - name: Install Python & Dependencies
         run: |
           apt-get update
-          apt-get install -y python3 python3-pip python3-venv git
+          apt-get install -y python3 python3-pip python3-venv python3-dev git
           python3 -m pip install --upgrade pip setuptools wheel
           
       - name: Install PyTorch (CUDA)
@@ -135,7 +136,7 @@ jobs:
       - name: Install Python & Dependencies
         run: |
           apt-get update
-          apt-get install -y python3 python3-pip python3-venv git
+          apt-get install -y python3 python3-pip python3-venv python3-dev git
           python3 -m pip install --upgrade pip setuptools wheel
           
       - name: Verify ROCm Installation

setup.py

@@ -51,20 +51,26 @@ def log(msg: str, level: str = "INFO") -> None:
 try:
     import torch
     from torch.utils.cpp_extension import CUDAExtension, BuildExtension, CUDA_HOME
-    TORCH_CUDA_AVAILABLE = torch.cuda.is_available() and (CUDA_HOME is not None)
+    
+    # FOR PRODUCTION/CI: Build if CUDA_HOME exists and CRAYON_FORCE_CUDA is set,
+    # even if no GPU is found on the build machine.
+    FORCE_CUDA = os.environ.get("CRAYON_FORCE_CUDA", "0") == "1"
+    TORCH_CUDA_AVAILABLE = (torch.cuda.is_available() or FORCE_CUDA) and (CUDA_HOME is not None)
+    
 except ImportError:
     TORCH_CUDA_AVAILABLE = False
     CUDAExtension = None
     BuildExtension = None
     CUDA_HOME = None
 
 # Detect ROCm
+FORCE_ROCM = os.environ.get("CRAYON_FORCE_ROCM", "0") == "1"
 ROCM_HOME = os.environ.get("ROCM_HOME", "/opt/rocm")
 HIPCC_PATH = os.path.join(ROCM_HOME, "bin", "hipcc")
-HAS_ROCM = os.path.exists(HIPCC_PATH)
+HAS_ROCM = (os.path.exists(HIPCC_PATH) or FORCE_ROCM)
 
 if HAS_ROCM:
-    log(f"ROCm detected at {ROCM_HOME}")
+    log(f"ROCm detected at {ROCM_HOME} (Forced={FORCE_ROCM})")
     log(f"hipcc found at {HIPCC_PATH}")
 else:
     log("ROCm not detected - skipping AMD backend")